Large-scale Optimization with 3D FEA Using Online Machine Learning Technique for Axial Flux Machine

Large-scale optimization of an axial flux machine using 3D FEA is performed using an online machine learning model. Computational optimization is a powerful tool and becoming indispensable for the design of high-performance machines. Since it requires massive FEAs, the computation cost tends to be high, therefore it is generally limited to 2D FEA. On the other hand, in pursuing higher power density, three-dimensional magnetic design needs to be taken into consideration, and computational optimization using 3D FEA is desirable. To investigate the viability of a 3D optimization, we conducted it for an axial flux machine with the help of a surrogate model to reduce the number of 3D FEA. The results showed that the use of the surrogate model effectively reduced the 3D FEA in the optimization, and highlighted that this method has the potential to make 3D FEA optimization a practical approach.